The AI era: from project interlocking to global efficiency

To maximize your chances of succeeding in school or other projects, lIA is increasingly part of the answer. Find out how to use it to manage or support your projects...

Before we look at what AI can do for you, let's take a look at some of its pitfalls

The number of people who start interacting with ChatGPT and think they've found the Holy Grail is impressive. But as with all tools, it's capable of doing some things well, others not at all. The risk is to make it do things it's not good at.

An amusing example given to me by my son Loïc, a civil engineer specializing in artificial intelligence, who uses AI in the field of marketing:

The design of AI models means they can be wrong.

The error in counting "a" may be due to a specific operating mode of the AI, such as :

• Word tokenization or segmentation: Language models like ChatGPT tokenize or segment words into smaller units for processing. Sometimes, this segmentation can lead to counting errors if part of the word is misinterpreted, or if characters are counted more than once.
  
  
• Answer generation: Generative AIs like ChatGPT don't literally count every character, but produce an answer based on what they predict to be the correct answer. This prediction can sometimes be wrong, especially if the model is based on patterns or associations of data rather than a true mathematical operation.

• Probability error: Answers are often based on probabilities. If the model has been trained on data where a similar error (such as counting 3 "a "s instead of 2 for a similar word) is frequent, it could reproduce this error.

Data dependency

AI relies on data sets to learn and function properly. If this data is biased, incomplete or unrepresentative, AI will produce results that are not good.
For example, in an educational context, if AI is trained on data reflecting existing biases, it could reinforce these biases in its recommendations or decisions. If you ask grok, Elon Musk's AI, to take a picture of a couple, it will almost always make couples representing diversity if you don't ask it anything specific like Caucasian or African-American because it's been trained that way :

Lack of empathy and context

Artificial intelligence lacks the innate ability to understand the emotional, cultural or contextual nuances of human interactions. This lack of empathy sometimes translates into inappropriate responses in situations requiring human judgment, such as psychological support or interpersonal conflict management. AI cannot interpret tone, subtleties of body language or complex social dynamics without detailed context.

Data security and confidentiality

There are a number of security issues that arise when using the web or connected media. AI is not exempt from this, but that's another aspect I won't go into beyond this commentary: AI systems often deal with sensitive information, particularly in the education sector where student and teacher data is at stake. The risks of data breaches, leaks of personal information, or even data manipulation by malicious actors, are major concerns. It's a good idea to think about security, as the use of AI can expose you to significant risks.

AI in education

Can AI help in schools? AI can revolutionize the way education is delivered and managed. In schools, it can :

• personalize learning paths by analyzing students' individual performance. This would make teaching tailored to the specific needs of each pupil or student.
  
  
• automate administrative tasks such as attendance management, grading or lesson planning. Detached from these tasks, this would leave teachers more time to concentrate on education and pedagogical support.
  
  
• become a helpful assistant for students with special needs, providing adaptive resources or immediate feedback.
  
  
• facilitate early detection of academic difficulties or well-being problems, enabling rapid, targeted intervention.

Source : Artificial intelligence in education

School IT systems

At the level of IT systems, AI would greatly simplify the management of tools used in schools. It can integrate and automate

• student management systems (SMS),
• e-learning platforms (LMS),
• administrative databases to provide a unified, accessible view of school information.

The benefits would be to reduce the complexity of navigating between different applications, reduce human error and increase operational efficiency.

For example, AI can optimize timetable management to avoid conflicts, automate the updating of student records and also predict maintenance requirements for IT equipment.

Coordination between school players

Artificial intelligence (AI) plays a pivotal role in coordinating the various players in the education system, strengthening the links between them for greater collective efficiency.

Facilitating communication and collaboration between students

With AI, we can imagine platforms that improve communication between teachers and students. A virtual place where exchanges can be personalized and instantaneous, such as intelligent messaging systems or interactive forums that facilitate the clarification of doubts, the exchange of educational resources or the organization of group projects.

For example, AI tools can offer communication channels tailored to students' learning styles, promote remote teamwork and why not provide personalized feedback on projects or homework.

Parents and educational players

It could also transform the relationship between parents and other school stakeholders, increasing transparency and commitment. It gives parents direct, real-time access to their children's academic progress, grades, absences and other performance indicators via personalized portals.

AI tools can automate administrative communications, such as reminders for parent meetings or report cards. It would even be possible to provide predictive analytics on educational needs or potential risks of school disengagement. This would create an environment where parents are more involved, informed and able to support their children's schooling.

Source : https://onlinedegrees.sandiego.edu/artificial-intelligence-education/

Optimizing resources

Savings can also be made by optimizing resources, both in terms of time management and the use of educational materials.

Time management

AI can rethink the way school time is managed by and within the school, enabling lessons and activities to be planned to take account not only of teachers' and classes' availability, but also of pupils' preferences and specific needs.

For example, AI algorithms can analyze historical data on absenteeism, academic performance and learner preferences to create optimized schedules. This includes allocating classes to maximize learning, minimize scheduling conflicts. AI-based tools can also dynamically adjust schedules to respond to unforeseen events or changes in resource availability, ensuring a seamless and efficient flow of teaching and activities.

Materials utilization

AI can also optimize the use of educational materials. It can analyze the use of books, digital tools, laboratory equipment, sports halls or classroom spaces to maximize efficiency. For example, AI systems can track the use of certain teaching aids to identify those that are under-utilized and redirect these resources to where they are most needed.

It could also forecast material requirements for courses, avoiding surpluses or shortages. AI can manage the maintenance and updating of technological equipment, ensuring that every tool is available and working, much like the company iCare does for industrial machinery, but on a different scale of course.

By monitoring wear and tear on physical resources and suggesting maintenance or renewal cycles, AI helps extend the useful life of materials and reduce long-term replacement costs. Every euro invested in educational resources would produce maximum educational value.

Helping to train better

Properly harnessed, AI can transform education by freeing teachers from administrative constraints and personalizing learning. By avoiding pitfalls and using it wisely, it offers a unique opportunity to refocus schools on their primary mission: to train and inspire.

---

[AI update on letter counting]

How many “a”s are there in the word “serpentaire”?

Chat GPT
There is only one “a” in the word “serpentaire.”

How do you count specific letters in a word?

I proceed in a mechanical and verifiable way:

    I consider the word as an ordered sequence of characters, as it is spelled.
    I go through each letter one by one, in order.
    I compare each letter to the target letter (for example, “a”).
    I increment a counter each time there is a match.
    At the end of the process, the counter gives the number of occurrences.

Applied to serpentaire:    

        s – e – r – p – e – n – t – a – i – r – e    

        → a single match with “a.”

This is therefore a count based strictly on the explicit spelling of the word, without phonetic interpretation or visual approximation.

Have you always done it this way, or have you learned to improve?

Chat GPT

More like the second option.

Initially, models like me did not “know” how to count letters in the strict sense. They produced plausible responses based on statistical regularities in language, which is often sufficient... but sometimes fails on mechanical tasks such as precise counting.

Over time, two things changed:

The training explicitly addressed this type of weakness.
Frequent errors (counting letters, words, parentheses, etc.) were identified, and the training reinforced more reliable strategies.

More procedural internal methods were encouraged.
Instead of “guessing” a linguistically probable outcome, I am encouraged to follow an explicit procedure (character-by-character scanning, step-by-step verification).

In other words, I have no personal memory or individual learning throughout the conversation, but the architecture and training have evolved to reduce these types of errors.
What you see here is not acquired experience, but a more disciplined version of reasoning.

---

See more articles by this author